Introduction and plan Domain of research
tarix 27.10.2017 ölçüsü 445 b. #16098
Distributed Artificial Intelligence and Ontology-based Systems for Knowledge Management. Domain of research: Distributed (formal) knowledge Distributed (artificial) intelligence PhD research: “D. A. I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D. A. I. and services” Multi-agent architecture & Semantic Web services Context-awareness for mobile accesses Knowledge-based privacy enforcement Wrap-up Web to humans
Web to machines...
Positioning (semantic Web) Internet & Web (shared infosphere for humans ) Intranets & intrawebsCurrent trend: reuse internet and web technologies Same advantages and same drawbacks Semantic Web (shared infosphere for machines ) XML : W3C standard for structuring data and documentsRDF(S) : W3C standard for metadata/semantic annotationsHyp : corporate memories as semantic intrawebs Positioning (ontology) Knowledge engineering & ontology (modularity, reuse) Assertional knowledge e.g., "Hugo wrote Notre-Dame de Paris "Ontological knowledge e.g., "Authors write books “Ontology: an explicit, partial account of the semantic structure encoding the rules that constrain our representation of realityHyp : ontology to support semantic intrawebPositioning (DAI) and context (CoMMA) Distributed artificial intelligenceMulti-agent systems: study design of artificial societies of intelligent agents Agent: clearly identifiable artificial entity; situated in an environment which it senses, reacts to and acts upon; self-control of its behavior ; social abilities to interact with other agents / humans; Multi-agent information systems: situated in & distributed over an information network Hyp European project CoMMA CoMMA Corporate Memory Management through Agents Atos-Origin, CSELT Telecom Italia, CSTB, INRIA, LIRMM, T-Nova Deutsche Telekom, University of Parma Two application and trial scenarios: Assist new employee integration Support technology monitoring activities Technical choices: Materialization of memory RDF(S) and its XML syntax (manipulated with CORESE )Exploitation of memory Multi-agent system and machine learning techniques (implemented with ) (implemented with WEKA )Positioning and pointers Dynamically integrating heterogeneous sources of information OBSERVER [Mena et al. , 1996] InfoSleuth [Nodine et al. , 1999] Carnot [Collet et al. , 1991] InfoMaster [Genesereth et al. , 1997] SIMS [Arens et al. , 1996] RETSINA [Decker & Sycara, 1997] Manifold [Kirk et al. ,1995] Assist the management of digital libraries SAIRE [Odubiyi et al. , 1997] UMDL [Weinstein et al. , 1999] Organizational knowledge management : Collaborative gathering, filtering and profiling CASMIR [Berney & Ferneley, 1999] Ricochet [Bothorel & Thomas, 1999] Mobile access, domain model, document classification KnowWeb [Dzbor et al. , 2000] Taxonomy of topics, profiling and push RICA [Aguirre et al. , 2000] Ontology and corporate memory: multiple ontologies FRODO [Van Elst & Abecker, 2001] semantic intraweb, ontology, user profiling You are here Functionalities: annotate, pull and push
Annotate documents (content awareness) RDF(S): annotated world for software in order to make inferences & help users exploit memory RDF = Resource Description Framework (annotation model) Annotation in RDF with XML syntax: Document rdf:about=“http://www-sop.inria.fr/aar.doc ”> Title>Annual activity report of ACACIA Author> Person rdf:about=“http://www.inria.fr/~rdieng /” /> Annotation in RDF with graph representation: Annotation schema (semantic awareness) RDF(S): annotated world for software in order to make inferences & help users exploit memory RDF = Resource Description Framework (annotation model) RDFS = RDF Schema (annotation vocabulary / ontology) Class rdf:ID=‘Entity’ /> Class rdf:ID=‘Document’ > subClassOf rdf:resource=‘#Entity ’ /> ... Property rdf:ID=‘Author ’> subPropertyOf rdf:resource=‘#Creator ’ /> domain rdf:resource=‘#Document ’ /> range rdf:resource=‘#Person ’ /> ... Annotate persons (context awareness) User modeling (extract of my user profile) Engineer rdf:about="http://www-sop.inria.fr/acacia/personnel/Fabien.Gandon /"> FamilyName>GANDON FirstName>Fabien BirthDate>31-07-1975 (...) HasForWorkInterest> MultiAgentSystemTopic rdf:about="http://www.inria.fr/acacia/comma#... (...) HasForPersonalInterest> HumanScienceTopic rdf:about="http://www.inria.fr/acacia/comma#Human... (...) Employee rdf:about="http://www-sop.inria.fr/acacia/personnel/Fabien.Gandon /"> HireDate>1999-11-02 EmployedBy> LocalOrganizationGroup rdf:about="http://www.ac-nice.fr/" /> EmploymentContract> Temporary/> HasForActivity> Research rdf:about="http://www.inria.fr/acacia/comma#Research "/> Idem for organization modeling (structures, relations...)Progression PhD research: “D.A.I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D.A.I. and services” Semantic Web services &Context-awareness Personal resource management & privacy enforcement Experimental results Wrap-up Methodological steps (1 & 2) Ontology building in five stepsStep 1 - Data collection and analysis Scenario-driven analysis: users’ scenario reports & grid Motivate data collection internal & external to organization Capture aspects conceptualization to assist scenarios Extract: Step 2 - Build a lexicon Capture terms and their definitions First intermediary representation of the ontology Constraint: one and only one occurrence of a definition Disambiguate terms, e.g.: COLLEAGUE n. (lat. collega ) someone who shares the same profession || one of a group of people who work together. Methodological steps in ontology engineering (3) Step 3 - Enriching lexicon structure Split concepts, properties and attributes into different tablesAugment with relevant semantic aspects (e.g. subsumption) Enrich, augment, refine, ... for both humans and machines Taxonomic skeleton: top-down / bottom-up / middle-outStep 4 - Script translating tables into RDFSAnalysis of the three levels present in RDF(S)
Methodological steps in ontology engineering (5) Step 5 - Factorizing knowledge (when needed)Declare algebraic properties of relations ( symmetric / transitive / reflexive relations) colleague(x,y) acquaintance(x,y) colleague(y,x) true acquaintance between two persons who work together. accointance entre deux personnes travaillant ensemble. colleague co-worker collegue collegue de travail Methodological steps in ontology engineering (5) Step 5 - Factorizing knowledge (when needed)Declare algebraic properties of relations ( symmetric / transitive / reflexive relations) No one generates all the instances of colleague by hand "I am a colleague of X because I work in the same group as X" (inference) Encode axiomatic knowledge in rules and definitions colleague(x,y) person(x) person(y) (z group(z) include(z,x) include(z,y)) IF (rule for sufficient condition) Group Include Person ?x Include Person ?y THEN Person ?x Colleague Person ?y Results Design stages: In CoMMA this method provided O'CoMMA 470 concepts (taxonomy depth = 13 subsumptions).79 relations (taxonomy depth = 2 subsumptions).715 terms in English and 699 in French .550 definitions in English and 547 in French .Resulting ontology: O'CoMMA Structure: Abstract top & middle layer for corporate memory: reusable Middle layer for domain: reusable in same domain Extension layer: usable but not reusable Reuse tested e.g., CSELT CSTB, APROBATIOM, KMP Progression PhD research: “D.A.I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D.A.I. and services” Semantic Web services &Context-awareness Personal resource management & privacy enforcement Experimental results Wrap-up A multi-agent architecture for CoMMA Problem = handle information distribution Handle naturally scattered information & knowledge Assist diffusion of captured information and knowledge Follow multi-agent paradigm: Artificial societies collaborating for global capitalizationArtificial individual intelligence , able to locally adaptStep 1 - Sub-societies identificationStarted from the tasks to be performed (provide ontology, manage annotations, manage users and matchmaking) Thus four sub-societies to handle these four tasks: Organizing resource-dedicated sub-societies Step 2 - Analyse possible organization structures: Roles and interactions
Zooming on annotation-dedicated sub-society
Annotation-dedicated sub-society (mediator)
Annotation-dedicated sub-society (archivist)
Interaction mediator-archivist (allocation) Interaction 1 - Annotation allocationPb: archives distributed all over organizationMediator & archivists discuss best archive for new annot. Contract-net (CfP, Proposal, Accept/Reject):Proposals: semantic distance new annotation - archive Content of an annotation or of an archive Simple annotation Article rdf:about="http://intranet/reports/R3029 "> Title>CfP UMTS Analysis Author> Person rdf:about="http://www.mycorp.com/~fab " /> Corresponding triples: Unstructured set of triples to describe content type Distance between an annotation and an archive
Building the distance Case 1 - literal values: lexicographical distanceCase 2 - concept types: minimum length of path between two types through least common super type DistH(Type1,Type2) = Min(GPath(Type1,LCST)+GPath(Type2,LCST)) GPath(,): number of edges through generalization links LCST: least common super type = shared characteristics Distance between two triples (conditional weighted sum) Semantic distance between types Distance = allocation criteria of contract-net "and the winner is..." the archivist with smallest distance Cluster annotations & specialize archivesImprove query solving & respect knowledge distribution Interactions mediator-archivist in solving a query
Overlap description
RDF query: tree structure
Decomposition
Decomposition
Overview of agent-based design Design stages: Result: Society providing ontology and corporate model Annotation-dedicated society Two trials at mid-project and project end in development spiral One public open-day demo : industrial & EU commissionUsability and usefulness recognized by users/publicProgression PhD research: “D.A.I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D.A.I. and services” Semantic Web services &Context-awareness Personal resource management & privacy enforcement Experimental results Wrap-up Web service to humans
Web service to machines
Positioning (semantic Web services) Internet & Web (shared service-sphere for humans ) Semantic Web (shared infosphere for machines ) Semantic Web Services (shared service-sphere for machines ) SOAP, XMLP, WSDL, DAMLS, UDDI (v2 v3) Hyp : Semantic web services for e-Business and Enterprise Applications IntegrationHyp : Ontology to support semantic Web servicesDistributed artificial intelligenceIntelligent Agent Intelligent Services Multi-agent systems Service interaction & composition Hyp: Multi-agent architectures for webs of servicesInterface without context awareness
Interface with context awareness
Positioning (mobile access to SW&S) Mobile (net)working Mobile networks: Phones / pagers PDAs, sub-notebooks Telecom services Web and online services Mobile accesses reduced & constrained interactions (interfaces, connectivity, cognitive workload) Schism: service availability & user availability Hyp : Context-awareness to support mobile access to semantic Webs and their servicesPositioning (context awareness) Context awareness State of the Art Application leveraging context awareness Active Badge [Want et al. , 92] ParcTab [Schilit, 95] Oxygen [Dertouzos, 99] GUIR [Hong & Llanday, 01] Aura [Garlan et al. , 02] Application dependent and heterogeneous Redundant and scattered Personal resources integration and unification Toolkits and widgets for wrapping [Dey et al. , 00] e-Wallet: awareness & privacy You are here Hyp : Semantic Web & Services to provide a unified secure interface to personal resources (e-Wallet)Context of research my Campus: a context-aware environment aimed at enhancing access to services for everyday campus life at Carnegie Mellon University (CMU)BBN, IBM, HP, Symbol and Boeing Air Force Research Laboratory (contract F30602-02-2-0035) Defense Advanced Research Project Agency (DARPA) (contract F30602-98-2-0135) Interactions with SONAT: user-aware notification (D.o.D.) CoSAR (I-X, KAoS/CoABS Grid): notification planning (AIAI) SWAP: Semantic Web and Peer-to-peer for KM (IST program) My focus: Semantic Web & Services, Context-awareness and Privacy, M.A.S. and Ontologies PDA & Wireless Network Agent roles: Platform manager User interaction manager Growing collection of task-specific agents e-Wallet manager Web resources Semantic Web services Semantic Web ontologies Semantic Web annotations Search engines Progression PhD research: “D.A.I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D.A.I. and services” Semantic Web services &Context-awareness Personal resource management & privacy enforcement Experimental results Wrap-up e-Wallet & the knowledge typology Secured unified semantic interface to access knowledge about the e-Wallet owner Static assertional knowledge: User’s static profile (name, SSN, position, etc. ) Static contextual knowledge (campus description, etc. ) fetched from outside as needed. Dynamic assertional knowledge: User’s dynamic profile (contextual preferences, etc. ) Context knowledge (location, current activity, weather, etc. ) fetched from outside as needed. Privacy: authorization & obfuscation (precision, lie, etc. ) Ontological knowledge, fetched from an outside repository of ontology at startup. Design of an e-Wallet Three-layer architecture Core knowledge: static & dynamic knowledge of user Service Layer: invoke external sources of knowledge: web services and personal resources Privacy layer: enforce privacy rules on external requests: access, obfuscation (precision, lies) Processes in the e-Wallet
e-Wallet core: JESS inference engine CLIPS language: Atoms, numbers, strings, functions, variables, Java reflection Ordered facts (not used here) Unordered facts (with templates // classes) Forward rules Queries: special kind of rule with no right-hand-side Declare backward chaining reactive rules using (do-backward-chaining …) function (template need-template) The RETE algorithm e-Wallet semantic engine RDF Triple model RDFS & OWL meta-model (e.g, symmetry of properties) e-Wallet semantic engine Ontologies: (e.g., declare man , location, etc. ) Annotations: (e.g., Fabien is in Smith Hall, etc. ) e-Wallet semantic engine Rules: (e.g., when in I am in a meeting I am busy) e-Wallet semantic engine Service and privacy layers Special types of triples Backward chaining reaction started by the query Privacy rules (clearance/revision) Service rules (dynamic access) Static migration rules Query Three steps: Query context assertion Query rule definition Body: request for authorized triples Head: storage & pretty printing function Rule execution Privacy rules
Service rules
Progression PhD research: “D.A.I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D.A.I. and services” Semantic Web services & Context-awareness Personal resource management & privacy enforcement Experimental results Wrap-up Prototype #1: proof of concept Mockup of multi-agent architecture Two task-specific agents Restaurant Concierge Agent RCA (location, weather, food, price, time) Message Filtering Agent (interests, current activity) + Jabber (Instant Messaging) Limited version of e-Wallet JSP pages to simulate User Interface Manager Three Web services Location tracking Calendar access Weather report Setting preferences
Using services
Experiment #1 with early prototype Before the experiment: Office for Human Research Protections approval Institutional Review Board certificate Selected group of 11 users, with a variety of profiles Trained the 11 users during a 2-hour session + material The 3-day experiment involved: Message filtering agent: 44 messages and 484 feedbacks Restaurant concierge agent: 28 recommendations Logs were generated for each one of these events. After the experiment: Users had to fill a survey ~1/2 hour Face-to-face de-briefing interviews ~15 minutes Statistics on the logs Experiment #1 extracts of feedback analysis
Experiment #1 most important results For the Restaurant Concierge, 12.5 % recommendations accepted thanks to context-awareness = 14.29% improvement. For the Messaging: filtering and routing based on static profile = too loose. finer knowledge about the users and finer filtering criteria; ~70% of the messages benefit from context-aware filtering and routing e.g. messages to be sent only when available or when the day is over Need critical mass of content, users and useful services Prototype #2: study of a useful service Focus: Find out about interesting events & Get information to people who want it Existing systems to distribute information? Design Computer Science Information Sc. Masters Electrical Eng. Experiment #2 H.C.I. study of a useful service Improving messaging for events: Tag each poster with a list of locations (situated) Display poster when user is close to it or explicitly interested in the topic (distribute) Peer-to-peer publishing (person-2-person) Integrate with calendar (remember) Stop publishing when outdated (maintenance) Under evaluation; good feedback Progression PhD research: “D.A.I. and knowledge” Semantic intrawebs for corporate memories Ontology engineering for semantic intrawebs Multi-agent architecture for semantic intrawebs Post-doc research: “D.A.I. and services” Semantic Web services &Context-awareness Personal resource management & privacy enforcement Experimental results Wrap-up Wrap-up: conclusion Ph.D. contributions Methodologies and tools to build ontology, semantic intraweb and associated multi-agent architecture A running prototype as a proof of concept for each point Both are reusable contributions Current extensions: Web scrappers & Semantic gateways Post-Doc contributions ( as a continuation) Design M.A.S. architecture for Semantic Web Services with a focus on mobile accesses Use of S.W. and ontologies to allow context-awareness Integration S.W. and ontologies in privacy enforcement Wrap-up: discussion Research ahead (remain at the state of the art) Ontology and Memory life cycles (collective-ware, coherence) Semantic Webs (semantic mapping, open/extra, community, security) Semantic Services and Agents (EAI, e-Sourcing, Grid, autonomic) Mobile & ubiquitous Semantic Web (connectivity, pervasiveness) Intelligent user interfaces (customization, awareness , ergonomics) Perfectly integrated with research of ACACIA & INRIA Acknowledgments ACACIA Laboratory - INRIA Sophia Antipolis Dr. Rose Dieng-Kuntz (Research Director – ACACIA project leader) Members of ACACIA team and logistics of INRIA Sophia ATOS-Origin, CSTB, Deutsch Telekom T-Nova, Italia Telecom, LIRMM and University of Parma IST Program (CoMMA project) School of computer science – Carnegie Mellon Uni. Prof. Norman M. Sadeh (Mobile Commerce Laboratory Director, SCS CMU, Free University of Amsterdam, European Commission) Members of my Campus team and logistic of ISRI BBN, IBM, HP, Symbol and Boeing IST Program (SWAP project) Air Force Research Laboratory (contract F30602-02-2-0035) Defense Advanced Research Project Agency (DARPA) (contract F30602-98-2-0135) Demo: CoMMA Ontology interface
CoMMA: profile-based retrieval
Demo: interface CoMMA
Demo: CoMMA annot. allocation interactions
Demo: CoMMA query solving interactions
Demo: InfoBridge virtual posters
Acknowledgments ACACIA Laboratory - INRIA Sophia Antipolis Dr. Rose Dieng-Kuntz (Research Director – ACACIA project leader) Members of ACACIA team and logistics of INRIA Sophia ATOS-Origin, CSTB, Deutsch Telekom T-Nova, Italia Telecom, LIRMM and University of Parma IST Program (CoMMA project) School of computer science – Carnegie Mellon Uni. Prof. Norman M. Sadeh (Mobile Commerce Laboratory Director, SCS CMU, Free University of Amsterdam, European Commission) Members of my Campus team and logistic of ISRI BBN, IBM, HP, Symbol and Boeing IST Program (SWAP project) Air Force Research Laboratory (contract F30602-02-2-0035) Defense Advanced Research Project Agency (DARPA) (contract F30602-98-2-0135) Dostları ilə paylaş:
